公开(公告)号：US20140244980A1

公开(公告)日：2014-08-28

申请号：US13775678

申请日：2013-02-25

Applicant: Yahoo! Inc.

Inventor： Masood Mortazavi

IPC: G06F9/30

CPC classification number: G06F9/50 ,  G06F11/30 ,  G06F11/3006 ,  G06F11/3447 ,  G06F11/3452 ,  G06F2209/501

Abstract: Method, system, and programs for dynamic control of a processing system having a plurality of tiers. Queue lengths of a plurality of nodes in one of the plurality of tiers are received. A control objective is received from a higher tier. One or more requests from the higher tier are processed by the plurality of nodes in the tier. A control model of the tier is computed based on the received queue lengths. One or more parameters of the control model are adjusted based on the received control objective. At least one control action is determined based on the control model and the control objective.

Abstract translation: 用于具有多个层的处理系统的动态控制的方法，系统和程序。 接收多个层之一中的多个节点的队列长度。 从较高层收到控制目标。 来自较高层的一个或多个请求由层中的多个节点处理。 基于接收到的队列长度计算层的控制模型。 基于接收到的控制目标来调整控制模型的一个或多个参数。 基于控制模型和控制目标确定至少一个控制动作。

公开(公告)号：US09104497B2

公开(公告)日：2015-08-11

申请号：US13671046

申请日：2012-11-07

Applicant: YAHOO! INC.

Inventor： Masood Mortazavi

IPC: G06F15/173 ,  G06F9/50 ,  H04L29/08

CPC classification number: G06F9/5083 ,  H04L29/08144

Abstract: Method, system, and programs for balancing work load in a distributed system. A plurality of resource units in the distributed system are ranked based a plurality of multi-dimensional load metrics determined from each of the plurality of resource units. The plurality of resource units are divided into a plurality of resource unit groups based on the ranking of the plurality of resource units. At least one pair of resource unit groups is formed. Each of the at least one pair of resource unit groups is assigned with a probability of load balancing action. For each pair of resource unit groups, a load balancing action is determined based on the assigned probability of load balancing action.

Abstract translation: 用于平衡分布式系统中工作负载的方法，系统和程序。 基于从多个资源单元中的每一个确定的多个多维负载度量，对分布式系统中的多个资源单元进行排名。 基于多个资源单元的排序，将多个资源单元划分为多个资源单元组。 至少形成一对资源单元组。 对至少一对资源单元组中的每一个分配有负载平衡动作的概率。 对于每对资源单元组，基于分配的负载平衡动作的概率来确定负载平衡动作。

公开(公告)号：US08886603B2

公开(公告)日：2014-11-11

申请号：US13864877

申请日：2013-04-17

Applicant: Yahoo! Inc.

Inventor： Masood Mortazavi ,  Jacob Y. Levy ,  Sahaya Andrews Albert

IPC: G06F17/30 ,  G06F11/34 ,  G06F11/20

CPC classification number: G06F17/30575 ,  G06F11/2094 ,  G06F11/3419 ,  G06F11/3495 ,  G06F17/30

Abstract: A ProbeSystem for one or more Replicas is disclosed. For each replica, the Probe System includes a ProbeHead that updates a Head field in a Target Record pertaining to the Replica and a ProbeEcho that updates an Echo field in the TargetRecord. The Head field includes a time stamp and an epoch value. The Echo field includes a latency value and an epoch value. The fields are such that, during reporting, the ProbeHead can analyze the TargetRecord for the replica to determine the latency between one or more paths between the Replica and other Replicas. The ProbeSystem has a peer-to-peer shared nothing architecture and applies to a large class of high-replica, distributed and high-scale cloud storage systems and services.

Abstract translation: 披露了一个或多个副本的ProbeSystem。 对于每个副本，Probe系统包括一个ProbeHead，用于更新与Replica有关的目标记录中的Head字段和更新TargetRecord中的Echo字段的ProbeEcho。 头字段包含时间戳和时期值。 回波场包括延迟值和时期值。 这些字段使得在报告期间，ProbeHead可以分析复制副本的TargetRecord，以确定副本和其他副本之间的一个或多个路径之间的延迟。 ProbeSystem具有点对点的无共享架构，适用于大型高复本，分布式和大规模云存储系统和服务。

公开(公告)号：US20140129715A1

公开(公告)日：2014-05-08

申请号：US13671046

申请日：2012-11-07

Applicant: YAHOO! INC.

Inventor： Masood Mortazavi

IPC: G06F15/173

CPC classification number: G06F9/5083 ,  H04L29/08144

Abstract: Method system, and programs for balancing work load in a distributed system. A plurality of resource units in the distributed system are ranked based a plurality of multi-dimensional load metrics determined from each of the plurality of resource units. The plurality of resource units are divided into a plurality of resource unit groups based on the ranking of the plurality of resource units. At least one pair of resource unit groups is formed. Each of the at least one pair of resource unit groups is assigned with a probability of load balancing action. For each pair of resource unit groups, a load balancing action is determined based on the assigned probability of load balancing action.

Abstract translation: 方法系统和用于平衡分布式系统中工作负载的程序。 基于从多个资源单元中的每一个确定的多个多维负载度量，对分布式系统中的多个资源单元进行排名。 基于多个资源单元的排序，将多个资源单元划分为多个资源单元组。 至少形成一对资源单元组。 对至少一对资源单元组中的每一个分配有负载平衡动作的概率。 对于每对资源单元组，基于分配的负载平衡动作的概率来确定负载平衡动作。

公开(公告)号：US20130232114A1

公开(公告)日：2013-09-05

申请号：US13864877

申请日：2013-04-17

Applicant: YAHOO! INC.

Inventor： Masood Mortazavi ,  Jacob Y. Levy ,  Sahaya Andrews Albert

IPC: G06F17/30

CPC classification number: G06F17/30575 ,  G06F11/2094 ,  G06F11/3419 ,  G06F11/3495 ,  G06F17/30

Abstract: A ProbeSystem for one or more Replicas is disclosed. For each replica, the Probe System includes a ProbeHead that updates a Head field in a Target Record pertaining to the Replica and a ProbeEcho that updates an Echo field in the TargetRecord. The Head field includes a time stamp and an epoch value. The Echo field includes a latency value and an epoch value. The fields are such that, during reporting, the ProbeHead can analyze the TargetRecord for the replica to determine the latency between one or more paths between the Replica and other Replicas. The ProbeSystem has a peer-to-peer shared nothing architecture and applies to a large class of high-replica, distributed and high-scale cloud storage systems and services.

Abstract translation: 披露了一个或多个副本的ProbeSystem。 对于每个副本，Probe系统包括一个ProbeHead，用于更新与Replica有关的目标记录中的Head字段和更新TargetRecord中的Echo字段的ProbeEcho。 头字段包含时间戳和时期值。 回波场包括延迟值和时期值。 这些字段使得在报告期间，ProbeHead可以分析复制副本的TargetRecord，以确定副本和其他副本之间的一个或多个路径之间的延迟。 ProbeSystem具有点对点的无共享架构，适用于大型高复本，分布式和大规模云存储系统和服务。